vp applied ai @moonpay, entendre (acquired) ex https://xitter.cf/t.co/WzNqDs4dyw
Joined January 2014
- Tweets3.2K
- Following895
- Followers937
- Likes35.1K
Pinned Tweet
@EntendreAI has joined @moonpay .
Last year everything accelerated. AI was starting to automate real financial knowledge work, and we felt it in everything we built.
Just as MoonPay is making payments, trading, and money movement headless, Entendre is making accounting and financial operations headless, across every financial touchpoint a company has, on and off chain. Their lead in stablecoins and moats across data and compliance make it the perfect place to build AI financial products at scale.
I owe so much to the people who made this possible: my team, who outbuilt companies ten times our size; our investors, who backed us before there was anything to back; and our customers, who trusted us not just with their financial health, but with their AI strategy.
Shout out to @ivanhodl & @KeithGrossman for seeing the same future we did. Can't wait to get to work.
We're hiring across the board. If you want to build financial intelligence at scale, DM me.
Today we're sharing that Entendre has been acquired by @MoonPay.
Three years ago we started with a simple conviction: the way money moves was changing fast, but the way it was tracked wasn't keeping up. Stablecoins were rewriting how businesses spend, earn, and transfer value, and accounting needed to move just as fast.
AI was the unlock. We built for the finance teams already operating at 9 to 10 figure volumes across multi-entity structures, wallets, exchanges, cards, subledgers, and bank accounts, fixing the actual work instead of forcing a rip-and-replace.
MoonPay is building the financial operating system for value: stablecoin-first, headless, agent-ready, institution-focused. It's the direction we were always headed.
For our customers: Entendre continues, now with more resources, faster deployments, and deeper integrations.
More to come. 🌑
kareem retweeted
In the words of @kvreem, we want our media company to be “ergonomic” to agents.
Another epic clip collection from @zdubs33
A video about agents, made with agents.
This is my second Stabledash Live "recap" video built with Opus 5.5.
For this test, I asked it to go longer, add more visuals, and focus on our agent conversations.
In February, when we launched the show, one episode talked about agents. By May, every episode did, and almost every one has since.
Opus 5.5 came up with "On the Record: Money for Machines" and split it into five chapters:
0:43 Built for humans
1:35 The customer is software
2:21 The rails
3:25 How big, really?
3:43 Money for machines
18 voices from 14 episodes, plus data it pulled itself from our archive, Visa, Allium and BlackRock.
I did spend more time on this one than the last video.
I'm still blown away by what it comes up with visually and the narratives it chooses to stitch together.
Jev is rapidly changing how we build agents for knowledge work. The endgame for accounting is a headless process: books that run themselves, reviewed by exception, not checked day in and day out.
How well can Jev (by @typesafeai) choose the right category for transactions that need accounting?
We tested it on our previous Crypto Accounting Benchmark (CAB), choosing from 2, 3, or 5 real accounts. Jev averaged 635 ms per successful call, with overall accuracy of 51.7%, 47.5%, and 48.3%, respectively.
With five candidates, keeping only cases with confidence ≥0.75 raised accuracy to 61.5%, covering 39 of 118 tasks.
The charts show both accuracy and how many cases remain at each confidence cutoff. The highest percentages come from the smallest samples.
kareem retweeted
How well can Jev (by @typesafeai) choose the right category for transactions that need accounting?
We tested it on our previous Crypto Accounting Benchmark (CAB), choosing from 2, 3, or 5 real accounts. Jev averaged 635 ms per successful call, with overall accuracy of 51.7%, 47.5%, and 48.3%, respectively.
With five candidates, keeping only cases with confidence ≥0.75 raised accuracy to 61.5%, covering 39 of 118 tasks.
The charts show both accuracy and how many cases remain at each confidence cutoff. The highest percentages come from the smallest samples.
kareem retweeted
MoonPay now powers deposits inside @moonshot's own interface.
@kvreem, VP of Applied AI at @moonpay, believes
"You should be able to move money, spend money, spin up cards, trade perps, meme coins, one-to-one stablecoins, whatever it ends up being. And you don't even know it's MoonPay."
xitter.cf/moonpay/status/2103498…
kareem retweeted
We really going to war with the fremen?
kareem retweeted
tricky thing is that it's an evaluation model, but maybe there's an interface that would make sence cc @fazxes
kareem retweeted
The best AI models get crypto accounting right only 56% of the time.
@EntendreAI published Crypto Accounting Bench this week, industry's first benchmark measuring how AI models perform on crypto-specific accounting tasks.
"That 56% accuracy, that's where Entendre plays."
"We can get people to that 100% because of the context that we have of your business, the data that we use to improve our agents."
"The models have a long way to go, especially when it comes to digital assets."
@kvreem VP of Applied AI @moonpay
xitter.cf/moonpay/status/2099861…
can AI models replace the accounting industry?
today we are publishing the Crypto Accounting Bench (CAB), the first open benchmark for AI crypto and stablecoin accounting, built by @EntendreAI
study up: entendre.ai/research/account…
kareem retweeted
all the cool kids are publishing an Accounting Bench this week
The underlying idea is that Jev is so fast and cheap that instead of one smart model call, you can fire 100-1,000 narrow structured decisions at once.
Jevons Paradox is the claim that making inference radically cheaper will increase total use, not merely lower cost.
OK after 24 hours of prototyping stuff with @typesafeai Jev, I'm realizing ~everything I built could've also just been shipped as Luna calls
Need to really lean into the name (Jevon's Paradox) to use-cases with 100x-1000x more throughput to find the new things this unlocks
kareem retweeted
"Rent your intelligence but own your data"
@kvreem shares the AI model strategy that every enterprise should adopt
"You should not be locked into one model.
"Let the best model win. There are advantages to both, especially if there's some sort of regulatory capture."
"But you as the enthusiast or as the person in charge of AI outcomes at a company, you should be Switzerland, and you should rent out your intelligence but own your context and own your data."
"That is the challenge a lot of enterprises face, that we actually help with on the back office side on @EntendreAI"
VP of Applied AI @moonpay on the live show today.
kareem retweeted
Live Thursday, September 17th, at 12:00 PM ET / 9:00 AM PT. The lineup:
- @kvreem, @EntendreAI / @moonpay
- @gaspardlezin, @subyhq
- @jjtoconnor, @realfi_co
- @InderpreetSingh, @D3inc
- @alexsrawitz, @crossmint
- @BTraidman, @HopInnovations
- @gaganmac, @arc / @circle
Drop questions below for the guests and we'll ask them live on the stream.
xitter.cf/i/broadcasts/1MJgNbPAj…
kareem retweeted
wow very interesting 🧐
MoonPay went from being known as the “buy crypto with your credit card” company to having one of the most stacked product lineups in all of crypto.
they now offer 20+ products across:
> on/off ramps
> virtual bank accounts
> fiat + crypto deposits
> global payouts
> stablecoin issuance
> wallets
> MPC/key management
> swaps
> crypto checkouts
> treasury management
> institutional trading
> reconciliation + accounting
> AI agent wallets
> agentic payments
a lot of this expansion has come from them going on a shopping spree and acquiring a dozen companies over the last year.
what @MoonPay gets especially right is that despite becoming a much bigger infrastructure company
they’ve never lost the consumer brand that made them relevant in the first place (due to their cracked social team)
A tier execution all around. has been beautiful to watch.